OK nerd, let's talk about it : a DVCS of course ; )
Mercurial here... But I may as well have gone the Git route. I used to use CVS and that was a world of pain, then moved to SVN and now it's Mercurial.
One caveat: you're a game designer right? So you may have big binary files like graphics, sounds, texture, etc. and if these changes often, then "stock" DVCSes can be quite sucky. So for example for Mercurial there's an extension (BFiles?) that makes working with big binary files more manageable.
But besides that caveat (for which there are solutions anyway), a DVCS is light-years ahead of non-decentralized version control system. You have to use one to realize how great they are. It takes some time to get use to it but it's really more than worth it.
We never thought about fixing bugs using what are now called "daggy fixes" because using a regular VCS such a workflow is just too painful. Using a DVCS introducing a daggy fix can be a matter of seconds (read more about daggy fixes here:
http://wiki.monotone.ca/DaggyFixes/ ). This alone is a night and day approach to bug-fixing and we couldn't go back to the old way of doing that (and reproducing that workflow using a VCS is doable but sssooo ssllooww).
Then push'ing and pull'ing over SSH is very, very efficient.
Regarding backups we have on-site on-line backups, on-site off-line backups, off-site on-line backups (on rented dedicated servers and on DropBox) and off-site off-line backups (that's what safes at banks are for ; ) All the off-site backups are basically the entire DVCS repos encrypted (there's no DVCS allowing encryption out of the box so sadly our simplest option for private codebases that we don't want to be "in the clear" on dedicated servers we're renting, or on DropBox, is to backup the entire repos and then encrypt it and then upload the encrypted file).
DVCSes allow for many different possible workflows and they're way, way, faster than non-decentralized VCSes. The regular VCS workflow can be re-created using DVCS but the inverse cannot.
Also DVCS are now quite big and do scale to the biggest codebases: the Linux codebase is using Git, the Java codebase is using Mercurial, etc. There's a reason these gigantic codebases worked on by huge teams were migrated to DVCSes...
Now I do still admin a dedicated server (because of all the coders I'm the one with the more Un*x skills

that's basically hosting more SVN repos than Mercurial ones but that's only because several developers still haven't spent the time learning how DVCSes work.
It's always the same thing: people still on CVS or SVN will say that they don't see why they'd need Git or Mercurial yet you'll have a very hard time finding someone who did try both DVCSes and VCSes telling you that SVN is better. Say for one developer who did try both SVN and Mercurial and that would prefer SVN, you'll find about a hundred that would prefer Mercurial.
That's my .2 commit on the issue and I won't enter in a VCS / DVCS war : )